This is an archive of the discontinued LLVM Phabricator instance.

[mlir][tensor] Do not use affine ops on non-perfect-tiling unpack cases.
ClosedPublic

Authored by hanchung on Feb 15 2023, 5:37 PM.

Details

Summary

Do not create an Affine ops for expanded size because the affine op is
too complicated which would hit an assertion in affine ops
simplification.

Diff Detail

Event Timeline

hanchung created this revision.Feb 15 2023, 5:37 PM
hanchung requested review of this revision.Feb 15 2023, 5:37 PM
Herald added a project: Restricted Project. · View Herald TranscriptFeb 15 2023, 5:37 PM

This is a similar fix to the issue we found in iree_linalg_ext.unpack version: https://github.com/iree-org/iree/issues/11607

It fixes infinite application of TileAndDistribute to unpack ops with dynamic tile sizes

mravishankar accepted this revision.Feb 16 2023, 11:28 PM

Maybe we should file a bug to fix the issue with simplification

This revision is now accepted and ready to land.Feb 16 2023, 11:28 PM
This revision was landed with ongoing or failed builds.Feb 17 2023, 11:08 AM
This revision was automatically updated to reflect the committed changes.